Superpower Daily
OpenAI Calls to Amend California’s SB 53 With Model Monitoring,...OVHcloud Cites AI Memory Demand as It Raises New Server Prices...University of Konstanz Finds AI Agents Coordinate Up to 1,000,...SemiAnalysis Says AgentX Drove 50-Plus Upstream Fixes for AI Ag...MiniMax H3 Opens Its Video Weights, but Not Where Many Develope...AI Data-Center Debt Moves Off Balance Sheets as Spending Could...OpenAI Calls to Amend California’s SB 53 With Model Monitoring,...OVHcloud Cites AI Memory Demand as It Raises New Server Prices...University of Konstanz Finds AI Agents Coordinate Up to 1,000,...SemiAnalysis Says AgentX Drove 50-Plus Upstream Fixes for AI Ag...MiniMax H3 Opens Its Video Weights, but Not Where Many Develope...AI Data-Center Debt Moves Off Balance Sheets as Spending Could...
Businesspublished

Every Built an AI Copy Editor From Its Editor in Chief’s 30,000 Edits

The internal tool is meant to extend Kate Lee’s editorial judgment across a growing company, not remove the need for expert judgment on difficult work.

By 3 min read
Every Built an AI Copy Editor From Its Editor in Chief’s 30,000 Edits
Sign in to save and react

Listen to this story

The audio brief

About 1:27
0:001:27
Read transcript
Every has turned about 30,000 of editor in chief Kate Lee’s past revisions into an internal AI copy editor that employees can use across the company. It can review articles, launch emails, landing pages, and other materials in what staff call a “Kate copy edit.” The idea is not to replace Lee on difficult work. It is to extend her standards and editorial taste to more documents, while freeing her time for higher-level decisions. CEO Dan Shipper said Every converted Lee’s earlier corrections into a prompt, tested it against documents the system had already seen, and refined it through repeated back-testing. The workflow also learns from what happens next. The agent can enter a Google Doc and suggest changes, while Lee’s later edits reveal what it missed. That feedback helps the team update the system, although Shipper says copy editing remains complicated and only partly automatable, even when it looks rule-based. The experiment sits inside a broader contradiction at Every. The company says AI writes essentially all of its code, yet its headcount doubled from roughly 15 people to about 30 in the past year. Shipper attributes that growth partly to new work created when AI output still needs expert adaptation. The constraint is that these advantages may not last. Every expects model capabilities to change quickly, sometimes forcing a product to be rebuilt within three to six months. The open question is whether an expert’s judgment can keep scaling as both the company and the underlying tools keep changing.

Story brief

3 key points

Every’s internal “Kate copy edit” turns roughly 30,000 revisions by editor in chief Kate Lee into a reusable agent for articles, launch emails, and landing pages. The system was developed through prompt refinement and back-testing, then connected to Google Docs so Lee’s later corrections can expose gaps. It illustrates a less obvious automation strategy: scaling one expert’s standards while creating more capacity...

  1. 01

    The agent was trained from about 30,000 historical edits and refined by testing its prompt against earlier documents.

  2. 02

    Employees can invoke a “Kate copy edit” across editorial and marketing materials, not just articles.

  3. 03

    Every’s headcount doubled from roughly 15 to 30 while the company pursued aggressive automation.

Every is automating aggressively while adding people. The publication and product studio says AI writes essentially all of its code, yet it grew from about 15 employees to roughly 30 over the past year. One expression of that approach is an internal copy-editing agent built from 30,000 historical edits by editor in chief Kate Lee.

An editor’s judgment becomes a company tool

CEO Dan Shipper said Every turned Lee’s prior revisions into a prompt, tested that prompt against earlier documents, and refined it through repeated back-testing. Employees can then invoke the internal system for a “Kate copy edit” on articles, landing pages, and other materials.

The target is broader than article copy. Shipper said Lee’s work includes launch emails and landing pages as well as editorial responsibilities. He described the agent as a way to distribute her particular standards and taste without requiring her to spend time on every document herself.

Better models made the workflow viable

Shipper said Every had pursued automation of parts of Lee’s copy-editing work since the GPT-3 era, without success for years. He attributed the more recent progress partly to improved instruction following, which let the company create a more useful prompt for varying situations.

Computer-use capabilities also changed the practical workflow, Shipper said. The agent can enter a Google Doc and make suggested changes; later edits by Lee are used to identify what it missed and update the system. Shipper’s assessment remains limited: the agent is not perfect, and copy editing is still complicated and not fully automatable despite appearing rules-based.

More automation, more expert work

Every’s growth complicates the simple story that automating tasks necessarily reduces staff. Shipper attributed the headcount increase partly to company growth and partly to work created when broadly capable AI output still needs expert adaptation to a specific problem. He said AI also enabled a single engineer to run a software product end to end at Every, something he said had not been practical at meaningful scale before.

That model has a separate business constraint. Every sells journalism alongside products including Cora, Sparkle, Spiral, and Monologue, but Shipper said the model makers it depends on can improve their systems and release competing application features. He said the company must be prepared to discard or remake products as capabilities change, sometimes on a three-to-six-month cycle.

For Every, the editing agent is therefore not simply a labor-saving feature. It is an attempt to preserve a scarce expert’s contribution while freeing that person for higher-level work. Whether that approach can keep producing useful editorial judgment as the company and its tools change remains a practical test of the system.

Sources

  1. platformer.newsThe website that created an AI clone of its editor in chief

Related coverage

Continue reading